TorchFort: An Online Deep Learning Interface for HPC programs on NVIDIA GPUs

Announcements
1

I think this new library may be interesting for those who want to do Deep Learning on NVIDIA GPUs directly from Fortran codes.

14 Likes
2

Thanks for posting this. There are two more similar projects (possibly less up-to-date) I’m aware of:

When I first started neural-fortran in the summer of 2018, Keras (high-level API over TensorFlow) was significantly more widely used than PyTorch, and so it was my main source of inspiration, especially user API-wise. However, PyTorch has since overtaken TensorFlow in popularity, and I’ve also heard a few notable people (e.g. George Hotz, creator of tinygrad; Chris Lattner, creator of LLVM, Swift, Mojo) praise PyTorch for its design. This motivates me to study PyTorch more to better understand ML frameworks and even get inspiration for neural-fortran where applicable.

TorchFort is a very cool name for a project, but I think it was a missed opportunity to call it torch.f90. :slight_smile:

6 Likes
3

TorchFort is a very cool name for a project, but I think it was a missed opportunity to call it torch.f90. :slight_smile:

How about FlowTorch => Fortran Learning Optimized with Torch :grin:

1 Like
4

Thanks for sharing the info, it looks interesting :flashlight: :bulb:

Btw, I’ve also come across this post (probably related to the above library).

Though I have no idea about the background, I feel that review processes are sometimes essentially “broken” (because of politics/preference/conflict of interest (?), etc, etc, …)

2 Likes
5

Indeed, definitely related and a very bizarre example of when badly managed peer-review can hurt researchers :frowning:

1 Like
6

Why not the easiest and simplest FoTorch?

7

Both might be the worst choices. :thinking: I could never imagine that naming would be a kind of art, too. Name is a symbol more than just a character string. :wink::grinning: